An Early-Onset Subgroup of Type 2 Diabetes: A Multigenerational, Prospective Analysis in the Framingham Heart Study

Type 2 diabetes (herein referred to as “diabetes”) is highly prevalent and manifests frequently at a younger age as well as at older age (1). Although it is known that diabetes confers substantial risk for cardiovascular disease (CVD), it remains unclear whether this risk significantly varies by age of diabetes onset (1–3). On the one hand, CVD risk may be more pronounced in earlier-onset diabetes, among persons with longer durations of exposure; on the other hand, cardiovascular risk may be greater in later-onset diabetes, among persons in whom age-related risk factors tend to aggregate (1). Clinical practice guidelines do not currently recommend distinguishing between patients whose diabetes developed earlier versus later in life, despite their potential etiologic as well as phenotypic differences, in the absence of any definitive evidence for varying prognosis according to age at onset. Data are emerging, however, to suggest that earlier-onset diabetes may represent a more aggressive form of disease, characterized by a much more rapid deterioration of the β-cell function with a more frequent need for insulin therapy and potentially higher rates of certain macrovascular and microvascular complications (1–3). Diabetes manifesting earlier in life might be more genetically determined, whereas diabetes presenting later in life may be more environmentally determined. If there exists a more aggressive subphenotype of diabetes that tends to present earlier versus later in life, is associated with greater risk for adverse outcomes, and is more likely to be heritable, then diabetes screening and disease prevention strategies could be targeted accordingly to identify this subset in the population at large. To test this hypothesis in the community- and family-based Framingham Heart Study (FHS), we examined 1) the long-term CVD risk associated with developing diabetes early versus late in the adult life course and 2) the extent to which early- versus late-onset diabetes in parents is associated with the risk of developing diabetes in offspring.

The study sample consists of participants in the first-generation (Original) and second-generation (Offspring) cohorts of the FHS. The Original cohort included 5,209 individuals enrolled from a sampling frame comprising two-thirds of the adult population aged 30–59 years in Framingham (in MA). These individuals enrolled into a prospective cohort study with biennial assessments starting in 1948. In the Offspring cohort, 5,124 children of the first-generation cohort were enrolled along with the children’s spouses, who were reassessed 8 years after the baseline examination (in 1971) and every 4 years thereafter. The details of the study design of these two cohorts have previously been published (4,5). For the mortality investigation, we only included the deceased participants from both the Original and Offspring FHS cohorts (n = 5,771).

The Boston Medical Center’s institutional review board approved all study protocols, and all participants provided informed consent.

At each examination, participants provided information on their medical history and underwent a physical examination and laboratory tests to assess cardiovascular risk (4,5). Nonfasting blood glucose was assessed at all Original cohort examinations and at the first two Offspring cohort examinations; in the latter cohort, fasting blood glucose was assessed at all subsequent examinations. We defined diabetes as presence of fasting blood glucose ≥126 mg/dL or nonfasting blood glucose ≥200 mg/dL, or treatment with glucose-lowering medications (6), at two consecutive examinations (to ensure the stability over time for a given glycemic phenotype). We defined disease onset as the first examination at which the criteria for diabetes was met (7), which was ascertained with use of all available plasma glucose data collected at serial examinations attended by the Original (1948 through 2005) and Offspring (1971 through 2008) cohort participants. We defined “early-onset” diabetes as diabetes diagnosed prior to age 55 years (8), given epidemiological data suggesting characteristics among persons with diabetes similar to characteristics among this age-group (9). We also selected age 55 years as the threshold for defining early-onset diabetes to optimize the number of individuals and, in turn, statistical power for analyzing individuals in our study who had died with ascertained causes of death prior to reaching this age threshold. We assumed that diagnosed diabetes was type 2 diabetes based on extremely low rates of type 1 diabetes in our cohort and the high prevalence of type 2 diabetes in the U.S. (7). All Original and Offspring cohort participants are under continuous surveillance for cardiovascular events and death; the process of adjudicating cardiovascular deaths has previously been described (10).

In a case-control design based on incident cases, we assessed the relations between age of diabetes onset and the cause of death (cardiovascular death [case] vs. noncardiovascular death [control]), using a pooled sample of Original and Offspring cohorts. We excluded individuals based on the following criteria: participant was still alive on 31 December 2014 (n = 85 Original cohort participants, n = 3,261 Offspring participants); participant did not attend any follow-up examinations (n = 135, n = 284, respectively); participant died at age <55 years without diabetes (n = 195, n = 94 [as for these individuals we could not exactly figure out their timing of diabetes onset, which could have been prior to their death or after had they lived longer]); participant had diabetes and was aged ≥55 years at baseline, precluding determination of early versus late onset (n = 17, n = 11); or participant had missing covariates (n = 478, n = 2). The final sample included n = 5,771 deceased participants (n = 4,299 Original and n = 1,472 Offspring).

For analyses of the cause of death, we categorized deceased participants as a case (i.e., CVD death) or a control (i.e., non-CVD death) subject. In our study sample, 1,822 individuals died of a cardiovascular cause (961 from coronary heart disease) and 3,949 died of noncardiovascular causes. We defined the exposure based on the age at onset of diabetes, with persons who died without ever developing diabetes serving as the referent group. We used multivariable logistic regression to relate case-versus-control status to age-group at onset of diabetes with adjustment for age at death, sex, smoking status, BMI, serum total cholesterol, use of statins, hypertension (defined as blood pressure [BP] ≥140/90 mmHg or taking antihypertensive medication [11]), use of antihypertensive medication, and duration of diabetes. The clinical covariates were drawn from the first examination cycle at which data were available; we also performed a sensitivity analysis using clinical covariates drawn from the last examination attended by the participant prior to death. We tested for a linear trend in odds ratios (ORs) across categories of age at onset of diabetes.

We investigated the extent to which age of diabetes onset in parents (early vs. late onset) was associated with risk for diabetes among offspring. We used Framingham Offspring participants who had both parents enrolled in the Original cohort (n = 2,470) with available data on diabetes status. We excluded Offspring who had diabetes at baseline (n = 25), who did not attend follow-up exams (n = 194), or who had missing covariates (n = 47). We also excluded Offspring who had any parents with indeterminate diabetes age of onset (n = 13) or any parents who did not attend any follow-up examinations (n = 68). The final sample for this analysis consisted of 2,123 Offspring participants.

For each Offspring participant, we categorized parental diabetes status as follows: 1) both parents free of diabetes, 2) one or both parents with late-onset diabetes, or 3) at least one parent with early-onset diabetes. We used Kaplan-Meier analyses to estimate cumulative incidence of diabetes in these categories. Because dates of diabetes onset were grouped based on FHS examination visit dates, we defined the date of diabetes onset in Offspring as the median date of each examination cycle. To assess the association between parental age of diabetes onset and risk of diabetes in Offspring (with adjustment for age, sex, BMI, and smoking status), we used a discrete-time logistic model. We tested for effect modification by sex by including a multiplicative interaction term in the model. For all analyses, we estimated robust SEs to account for correlation between siblings in the offspring (12).

Additionally, we investigated whether younger compared with older “age-at-risk” in offspring might influence the association between parental diabetes status and incidence of diabetes in offspring by conducting analyses using two age-groups at risk (<55 and ≥55 years). We then determined age-at-risk–specific risk estimates by incorporating an interaction term between age-at-risk and parental diabetes status in models with adjustment for the same covariates listed above (with age replaced by year of birth). Among offspring who developed diabetes (n = 164) and had one or more parents with diabetes, we also examined the extent to which age of onset of diabetes in parents was associated with the age of onset of the condition in their offspring. For these analyses, we used multivariable-adjusted discrete-time logistic models fit with generalized estimating equations to account for sibling correlations. We defined parental age of onset as the lower of the two available ages if both parents had developed diabetes.

All analyses were performed with SAS software, version 9.4 (SAS Institute, Cary, NC). We considered a two-sided P < 0.05 as statistically significant.

The characteristics of our study participants are shown in Table 1 and in Supplementary Table 1. Participants who died of CVD, or specifically coronary heart disease, had experienced diabetes onset at a younger age than those who died of noncardiovascular causes. Individuals who died of cardiovascular causes were also less likely to be women and more likely to have a high blood glucose, high serum total cholesterol, and hypertension. Participants with earlier age of diabetes onset tended to have higher BMI, higher total cholesterol, lower HDL cholesterol, and greater prevalence of hypertension at baseline. Among individuals with early-onset diabetes, the mean duration of diabetes tended to be shorter among those who died of CVD than among those who died of a noncardiovascular cause (Supplementary Table 2).

Table 2 displays the risk estimates for cardiovascular versus noncardiovascular deaths by age of onset of diabetes groups. The trends of decreasing age of onset of diabetes in relation to increasing odds of cardiovascular and coronary mortality were statistically significant (P < 0.05 for all). When compared with individuals who never developed diabetes, people with onset of diabetes prior to age 55 years were associated with 1.81-fold greater odds (95% CI 1.10–2.97; P = 0.02) of cardiovascular death. In contrast, onset of diabetes at age ≥65 years did not confer a statistically significant higher risk of cardiovascular death (OR 1.18, 95% CI 0.96–1.46; P = 0.12) (Fig. 1). Onset of diabetes before age 55 years (OR 1.75, 95% CI 0.96–3.21; P = 0.07) or at age ≥65 years (OR 1.09, 95% CI 0.85–1.38; P = 0.57) did not significantly increase the risk of coronary death (Table 2), whereas onset of diabetes at age 55–64 years increased this risk by 1.59-fold (95% CI 1.02–2.49). Onset of diabetes before the age of 55 years was associated with a 2.15-fold greater odds (95% CI 0.35–13.17; P = 0.41) of stroke death, but onset of diabetes at ≥65 years age did not significantly increase risk (OR 1.05, 95% CI 0.43–2.56; P = 0.91) (Supplementary Table 3).

We observed a trend for increasing odds of stroke death with decreasing age of onset of diabetes (P < 0.01 for trend) (Supplementary Table 4); this trend was significant but appeared less obvious than that observed for cardiovascular death and coronary death. In sensitivity analyses including covariates drawn from the last attended examination, results were similar to those in the main analyses (Supplementary Table 5). Given temporal trends in smoking status in our study sample that were consistent with trends reported previously, we performed analyses adjusting for smoking status both at the first examination and at the last examination prior to death and observed similar results (Table 2 and Supplementary Table 5). Results were also similar with additional adjustment for socioeconomic status, based on information regarding education and occupation that was collected at select examinations (Supplementary Table 5). We tested for effect modification of the association of age of diabetes onset and risk of CVD or CHD death through the inclusion of an average BMI (over the study period) interaction term in adjusted models for each of the outcomes. We observed no heterogeneity of effect by BMI.

The baseline characteristics of the sample (n = 2,123) used to assess the relation of age of onset of diabetes in parents to the risk of diabetes in their offspring are shown in Supplementary Table 6. Across the categories of parental age of diabetes onset, we observed differences in age, fasting blood glucose, BMI, HDL cholesterol, systolic BP, and diastolic BP. Over the follow-up period (mean ± SD 32 ± 10 years), 164 Offspring participants (7.7%) developed diabetes. The incidence of diabetes was higher for those with parents who had early-onset versus those with late-onset diabetes or those who remained free of diabetes (log-rank P < 0.001) (Fig. 2). Compared with having parents without diabetes, having at least one parent with early-onset diabetes significantly increased the risk of developing diabetes, more so than having one or more parents with late-onset diabetes (Table 3). Having even just one parent with early-onset diabetes increased the risk of developing diabetes 3.24-fold (95% CI 1.73–6.07; P = 0.0003) after adjustment for covariates. We investigated how age-at-risk affects the association between parental diabetes status and incident diabetes risk by dividing the follow-up into two age-at-risk bands (<55 and ≥55 years). Parental diabetes status conferred similar hazard for incident diabetes in both age-at-risk bands (Supplementary Table 7). There was no interaction by sex with respect to the association of parental diabetes status with risk of developing diabetes in offspring (P = 0.45). The use of an alternative definition of early-onset diabetes as manifesting at age <50 years (instead of <55 years) yielded similar results (Supplementary Table 8).

We conducted a multigenerational study of diabetes and cardiovascular mortality risk in the FHS and observed that a potentially important subset of diabetes may be defined based on the age of onset of diabetes. Specifically, we observed that when diabetes occurs prior to age 55 years, which is increasingly evident among adults who develop the disease (13), there is a significantly greater lifetime risk for CVD compared with diabetes that manifests at a later age. We also noted that individuals with earlier onset of diabetes were more likely than those with a later onset of diabetes to have offspring who develop diabetes. Taken together, these results provide evidence for a subgroup of diabetes that appears to manifest earlier in adulthood to be more severe in its association with cardiovascular death and to cluster across generations within families.

It is well known that diabetes is a heterogeneous disorder in terms of its natural history. Individuals with diabetes can vary widely in their disease course and complications—and this variation poses ongoing clinical challenges for diagnosing and managing affected persons (3,14). Thus, as part of efforts to identify higher-risk disease subgroups amid heterogeneity, prior studies have assessed the effects of earlier versus later onset of diabetes on the CVD risk posed, and the results of these studies have been mixed. On the one hand, studies of hospital-based or select samples with predominantly cross-sectional exposure data have suggested a higher burden of CVD risk factors (15,16) and a greater risk for CHD or stroke among patients with earlier- versus later-onset diabetes (1,17–20). On the other hand, some studies have suggested either a lower risk or no difference in risk for macrovascular complications among persons with earlier- versus later-onset diabetes; these studies, however, have relied mainly on patient or physician report of diabetes age of onset (21,22). A recent study by Sattar et al. (23) showed that individuals aged ≤40 years at diagnosis of type 2 diabetes had the highest excess risk (relative to control subjects) for cardiovascular-related mortality, and coronary heart disease, compared with individuals who were older at diagnosis. Although using data from medical diagnosis codes and a shorter duration of follow-up (∼5.6 years), this study also found that earlier-onset diabetes was associated with greater risk even after accounting for diabetes duration (23). Extending from these results, we defined early- versus late-onset diabetes using objective laboratory and clinical data collected from serial examinations in a community-based cohort with >30 years of prospective follow-up. The multigenerational FHS design allowed not only for a comprehensive assessment of long-term CVD risk in relation to age of diabetes onset but also for examining the correlates of early-onset diabetes; of prior studies reporting on the familial risk of diabetes, none have specifically examined risk for an early-onset disease phenotype in relation to the parental diabetes status.

Our study was motivated by a persistent uncertainty regarding the relative risk conferred by onset of diabetes earlier versus later in life. Although diabetes onset in older age is often associated with presence of additional cardiovascular risk factors that can substantially increase cardiovascular risk when combined, diabetes onset at a younger age may represent a more aggressive subgroup that confers greater risk even after the longitudinal effects of disease duration are accounted for. In our study cohorts, under surveillance for outcomes over up to six decades of follow-up, we observed that objectively assessed diabetes onset before age 55 years was associated with a >1.7-fold greater odds of CVD death, particularly coronary death, compared with onset of diabetes at an older age. Furthermore, presence of earlier-onset diabetes in parents was associated with a 3.5-fold greater odds of developing diabetes in their offspring. Conversely, onset of diabetes at an older age was not significantly associated with greater risk for CVD death, compared with persistent no diabetes status, although it was associated with a 2.3-fold odds of developing diabetes in offspring. Our main findings, indicating that cardiovascular mortality risk is actually lower in persons with diabetes onset at older compared with younger age, are consistent with results from a large multiethnic study that found lower hazard for cardiovascular death among individuals with diabetes aged <60 years compared with those aged ≥75 years (24). The observation of a lower risk of CVD-related death among those aged ≥65 years compared with younger participants, while potentially unexpected, is also congruent with recent U.S. national trends data suggesting a much higher decrease in CVD-related death among individuals with diabetes aged ≥65 years than in younger age-groups over the last 20 years (25,26).

There are several possible explanations for our main findings. The difference in CVD risk seen between persons with earlier- versus later-onset diabetes is at least partly attributable to different durations of exposure and the fact that younger compared with older persons have a longer time window during which CVD may develop. However, similar to Sattar et al. (23), we observed that adjusting for differences in disease duration did not substantially alter the main results of our longer-term study. Furthermore, consistent with Sattar et al. and investigators of smaller studies of select populations (e.g., American Indians [27], Asian Indians [28], and Iranians [29]), we also found that individuals with earlier- versus later-onset diabetes appear to have more pronounced clinical features of metabolic disease such as greater obesity, adiposity, lipid abnormalities, and dysglycemia. Taken together, and despite the interrelated effects of diabetes age of onset and disease duration, the evidence to date suggests that earlier- versus later-onset disease likely represents a distinct subgroup phenotype of type 2 diabetes that confers risks above and beyond the effects of prolonged disease duration. In effect, early-onset diabetes may be a phenotype distinct from late-onset diabetes, rather than the same disease simply manifesting at a different point in life. The greater degree of association between early-onset diabetes status than late-onset diabetes status in parents and diabetes risk in offspring supports this notion. Indeed, familial aggregation of subclinical CVD traits appears more prominently in some but not all persons with diabetes (30,31); risk conferred by certain genetic variants (e.g., 9p21, 1q25) for coronary disease is also more pronounced in some but not all persons with diabetes (32,33). If age of onset of diabetes can identify an especially high-risk subgroup of diabetes that confers a higher risk of CVD and evidence of clustering within families, age of diabetes onset could represent a potentially useful tool for prioritizing CVD prevention measures and identifying family members with potentially shared diabetes risk.

Some limitations of this study merit consideration. With respect to characterizing a potentially distinct subgroup of diabetes, we did not analyze repeated measures of insulin resistance, glycemic control, microvascular complications, or macrovascular complications other than overt CVD mortality events; the extent to which these measures may also vary between individuals, with earlier versus later onset of diabetes warrants further investigation. With respect to analysis of specific CVD mortality events, the number of deaths attributable to stroke was relatively small in our study; thus, less obvious trends observed for stroke death are likely due to limited statistical power for these analyses. Given limited statistical power in our sample for analyzing outcomes related to early-onset diabetes defined using age thresholds <55 years, studies with larger numbers of younger individuals are needed for further investigations. Our assessment of diabetes in the Original cohort was solely based on nonfasting glucose measures and lack of consistently available data from oral glucose testing; thus, some cases of early-onset diabetes may have been misclassified as late-onset diabetes, which would have biased our results to the null. Given increased awareness among providers and patients, the time from diabetes onset to clinical diagnosis has likely shortened over time, although our longitudinal data offer insufficient temporal resolution to investigate the potential implications of this trend for individuals with either early- or late-onset diabetes. Similarly, our data do not include longitudinal insulin resistance or detailed assessments that might otherwise allow for investigating time lags between mechanistic onset of and overt manifestation of diabetes. We also did not account for the changes over time in the pharmacological approach to the primary and secondary prevention of CVD (including the management of cardiometabolic risk factors), which may have affected the outcomes in our study. Our study included individuals of predominantly European ancestry; thus, the generalizability of results to other racial/ethnic groups remains unknown.

We identified a subgroup of diabetes, objectively defined based on a relatively early age of onset that confers an increased long-term risk for CVD in affected individuals and a substantial shared risk for diabetes across generations. These findings have implications for identifying propensity for diabetes in family members and for prioritizing efforts to reduce CVD risk in persons with prevalent diabetes—particularly younger individuals, in whom rates of cardioprotective medication use remain suboptimal (34). In effect, approaches to clinical diagnosis and management of diabetes-related risk could potentially benefit from collection of not only general personal and family history of diabetes but also information specifically on diabetes age of onset. However, further work is needed to confirm our findings and to determine the extent to which more targeted interventions aimed at reducing CVD risk among individuals with early-onset diabetes could lead to improved outcomes in this particularly high-risk population.

Acknowledgments. The authors thank the staff and participants of the FHS.

Funding. This work was supported by the National Heart, Lung, and Blood Institute’s FHS (contracts N01HC25195, HHSN268201500001I, and 75N92019D00031) and the following National Institutes of Health grants: T32 HL125232 (J.B.E.-T.), R01HL093328 (R.S.V.), R01HL107385 (R.S.V.), R01HL126136 (R.S.V.), R00HL107642 (S.C.), R01HL131532 (S.C.), and R01HL134168 (S.C.).

The views expressed in this manuscript are those of the authors and do not necessarily represent the views of the National Heart, Lung, and Blood Institute; the National Institutes of Health; or the U.S. Department of Health and Human Services.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.B.E.-T., M.G.L., T.J.N., and S.C. contributed to the study concept and design. J.B.E.-T., T.J.N., E.L.M., M.H., M.J., R.S.V., M.G.L., and S.C. contributed to acquisition, analysis, or interpretation of data. J.B.E.-T. and S.C. contributed to drafting of the manuscript. T.J.N., E.L.M., M.H., M.J., R.S.V., M.G.L., and S.C. contributed to critical revision of the manuscript for important intellectual content. E.L.M., M.H., and M.G.L. contributed to statistical analysis. R.S.V. and S.C. obtained funding. R.S.V., M.G.L., and S.C. contributed to administrative, technical, or material support. S.C. supervised the study. J.B.E.-T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.